Model versions
Each Abstractus tool has a numbered model version. When the screening process changes, the version number is updated. Every screening run records the version used so you can identify how its results were produced.
Current models
| Product | Version | Benchmark | Price |
|---|---|---|---|
| Abstract screening | Version 2 (v2.0.0) | 94.9% median recall · 87.7% median workload reduction | 1 credit per abstract |
| Full-text screening | Version 1 (v1.0.0) | 100% median recall · 97.3% median specificity | 2 credits per page |
| Data extraction | Version 1 (v1.0.0) | No public benchmark yet | 4 credits per page |
| Pharmacovigilance | Version 1 Preview (v1.0.0-preview) | No public benchmark yet | 4 credits per page |
Recall is the percentage of relevant papers included by the screening. Workload reduction is the percentage of all papers excluded. Each screening job has a 50-credit minimum, and each extraction job 100.
Abstract screening
Evaluates each abstract against the inclusion and exclusion criteria individually, then uses those assessments to make the final screening decision.
- Organizes your criteria into rules by theme and evaluates each rule separately.
- Provides a short reason for each decision by default. You can request detailed reasons for all papers or only included papers.
- Each excluded paper is grouped under its most relevant exclusion criterion, shown in the Breakdown tab and in downloads.
- Benchmark
- 94.9% median recall · 87.7% median workload reduction
- Price
- 1 credit per abstract
18 public SYNERGY reviews · 56,384 abstracts · Same reviews and method as the version 1 benchmark; write-up to follow.
Previous versions (1)
Converted your criteria into screening rules and refined those rules using a sample of papers. It then screened every abstract against the refined rules and provided a reason for each decision.
- Benchmark
- 94.4% median recall · 84.0% median workload reduction
- Price
- 1 credit per abstract
18 public SYNERGY reviews · 56,384 abstracts · Abstract screening benchmark
Full-text screening
Evaluates each full-text PDF against your criteria and provides a screening decision with a reason.
- Benchmark
- 100% median recall · 97.3% median specificity
- Price
- 2 credits per page
11 public reviews · 3,338 full texts · Full-text screening benchmark
Data extraction
Extracts the information specified in your form from each paper and compiles it into a data table and report. Each value links to its supporting quote.
- Benchmark
- No public benchmark yet
- Price
- 4 credits per page
Pharmacovigilance
Reads each paper for one drug, finds every patient with an adverse event, and decides for each patient whether it is a valid case and whether it meets the EU and US 15-day reporting rules. Each fact links to the passage it came from.
- Screens each paper first, and gives a short reason for every paper it rules out.
- Checks expectedness against the drug's dated US and EU labels where we hold them.
- Runs once on your PDFs, or every week as a monitor across your chosen sources.
- Benchmark
- No public benchmark yet
- Price
- 4 credits per page